運用資料探勘方法預測雜誌銷售量

Abstract

連鎖零售業銷售的商品包含生鮮食品、日用品、玩具及文化出版品等，並定時將這些商品配送至全台各門市銷售，而每家門市因地理位置、人口結構、區域屬性、天氣、季節等外在環境變因的影響，各家門市的需求量不一，因此，如何準確的預測門市商品需求量，成為零售業者為增加獲利所必需研究的課題。 本研究將針對女性時尚雜誌月刊進行門市需求量的預測，主要是因為觀察到女性時尚雜誌的銷售曲線有明顯的季節性循環，但現行的需求量預估採簡單移動平均法則很容易低估市場需求，因此本研究以自我迴歸樹時間序列及灰預測兩種方法進行門市銷售量預估，並以平均絕對誤差(MAE)衡量其預測準確度。 研究結果最後顯示，當門市銷售量與季節性指數相關性較高時，預測效果以多變數自我迴歸樹時間序列方法優於單變數自我迴歸樹時間序列方法及灰預測方法，但當門市與季節性指數相關性較低時，以灰預測方法優於單變數自我迴歸樹及多變數自我迴歸樹時間序列方法。

Chain of retail sells fresh food, groceries, toys, publications and so on. They distribute these goods to the franchises regularly. The demand of each franchise is different because of their geographic, population structure, residential area, weather, seasonal changes and extrinsic environmental impact. Therefore, how to accurately forecast demand for the amount of goods to increase profit is one of the most important business issues. The research focuses on the prediction about the retail demand of female fashion magazine. It is due to an observation that there is a seasonal cycle in sales curve of female fashion magazine. However, the chain of retail uses simple average in sales amount prediction is easy to undervalue the market demand. In this research, we adopt Autoregressive Tree Time Series method and Grey Prediction method to construct sales prediction model, and take the MAE as criteria to evaluate the forecasting competences of different models. According to the empirical results, if the sales series and the seasonal index have higher correlation, then the prediction accuracy of multivariable ART (Autoregressive Tree) is better than those of Univariable ART and Grey Prediction methods; otherwise, the prediction accuracy of Grey Prediction method is better than those of Univariable ART and multivariable ART methods.